ZynoviqClean files in your schema, under a licence that says exactly who may do what.
There are three decisions when buying data from us: the format it arrives in, how it is made (custom or ready-made), and who else may use it (shared or exclusive). This page covers all three, and the reference packs we make for agents that cannot be retrained.
See a sample dataset and the schema → · Browse the catalogue →
File formats
| Format | Typical use |
|---|---|
| JSONL | Training and preference data (one record per line) |
| Parquet | Large datasets and analytics pipelines |
| CSV | Evaluation results and simple tables |
| JSON + Markdown or PDF | Reference packs |
We deliver in your schema when you specify one; otherwise in our documented standard schema, shared with every sample. The standard schema, field by field.
Every delivery includes
| File | Contents |
|---|---|
| data files | The records themselves |
| provenance.json | Per record: version, valid-as-of date, anonymous contributor ids, reviewer agreement, adjudication status |
| quality_report.pdf | Counts by topic and difficulty, agreement rates, how certification was done |
| licence.json | Licence type, permitted uses, term, licensee, copy identifier |
| changelog.md | What changed since the previous version |
How we deliver
| Method | When |
|---|---|
| Secure download by expiring link, with download logs | Most deliveries |
| Direct transfer to your cloud storage: we push to your bucket, then remove our temporary copy | Large deliveries to AI labs |
| Batches on a schedule | Missions delivered in stages, and expert case subscriptions |
We never deliver licensed data through public code repositories. Data is stored and delivered from Indian cloud regions. Data processing · Trust.
Each copy is unique. Every licensee's copy carries a unique identifier and small, harmless variations, so any leaked copy can be traced.
Updates. When the law changes, you receive the updated records as a new version with a change note. You choose whether to retrain. Evidence and currency.
Custom or ready-made
| Custom (a Data Mission) | Ready-made (from the catalogue) | |
|---|---|---|
| What it is | Built to your specification: domain, topics, period, difficulty mix, format, schema | A dataset we have already built and certified |
| Speed | Weeks, depending on size; a typical first mission takes five to seven weeks | Available when production finishes; see the status on each catalogue entry |
| Fit | Exactly your need | Close to your need; may be extended |
| How it is priced | Fixed price against the approved specification, quoted | Per licence, quoted; usually lower because the cost is shared |
Shared, exclusive or owned
| Licence | What it means | Relative price |
|---|---|---|
| Shared (non-exclusive) | Others may license the same data | Lowest |
| Exclusive for a period | Only you, for an agreed time or field | Higher |
| Owned by you | Built only for you and transferred to you | Highest |
Why exclusivity costs more. Making a certified record takes real professional time: several independent solvers, review, adjudication and a legal check. When data is shared, that cost is spread across several buyers, so each pays less. When it is exclusive, one buyer carries the whole cost and we cannot reuse the work. Exclusivity is therefore priced separately, because it forecloses reuse.
Which to choose
| If you… | Consider |
|---|---|
| Need a fast start on common topics | Ready-made, shared |
| Need your own format or unusual topics | Custom, shared |
| Compete on this exact capability | Custom, exclusive for a period |
| Must own the data outright (policy or contract) | Custom, owned |
Whichever you choose: certified answers, full provenance, a quality report, a clear licence, and change notes when the law moves. Licences state who may access, redistribute, publish or train on the data. Full terms: Licence types · Data rights.
Reference packs: a rulebook your agent can look up
Many organisations cannot retrain their AI: an SAP-delivered agent, a packaged assistant, a vendor's product. What they can do is give it better material to look up while it works. A reference pack is that material: certified rules and worked examples, written for an AI to find and use.
| Part | Example |
|---|---|
| Rule summaries | "Credit on food and beverages is blocked, except where the employer must provide them by law" |
| Conditions and exceptions | Each exception stated separately, with its conditions |
| Worked examples | Short synthetic cases showing the rule applied, including the common mistake |
| Decision guides | Step-by-step checks, for example: is this a notified reverse-charge service? |
| Citations and dates | Authority for every statement; valid-from and valid-until dates |
How it is used. Your team loads the pack into the source your agent searches: a knowledge base, a document store or your platform's grounding feature. Each pack comes with a short loading guide for SAP and other common platforms.
Kept current. Reference packs are updated when the law changes, with a change note showing exactly what moved. Out-of-date material is the fastest way to make an agent confidently wrong.
Packs in production or planned
| Pack | Status |
|---|---|
| GST — input tax credit and blocked credits | In production |
| GST — reverse charge | Planned |
| GST — place of supply | Planned |
| TDS on vendor payments | Planned |
Test it, do not assume it. After loading a pack, re-test the agent to measure what improved. We describe the result honestly as a before-and-after measurement. Improvement data.
Not tax advice. Reference packs are material for AI systems, not advice on any organisation's own tax position. Terms of use.
SAP is a trademark of its owner. Zynoviq is not affiliated with or endorsed by it.
Formats, delivery and licences — FAQ
What files come with every delivery?+
The data files, a provenance file, a quality report, a machine-readable licence and a change log. We share a small sample in your schema before any full delivery, so your pipeline is tested early.
Can you deliver data through GitHub?+
No. We never deliver licensed data through code repositories, only by expiring secure download link or by direct transfer to your cloud storage.
Should we buy custom or ready-made data?+
Ready-made is faster and usually costs less for common topics; custom fits unusual topics or your own schema exactly. Many labs start with a ready-made pack and commission a Data Mission for the gaps.
Is an exclusive licence permanent?+
No. Exclusivity lasts for an agreed period or field, and is priced separately because it forecloses reuse. Only custom work can be transferred to you outright. Licence types.
Why is each licensed copy unique?+
Every licensee's copy carries an identifier and small, harmless variations, so any leaked copy can be traced. Data rights.